Sie sind auf Seite 1von 9

Serial Murders Spatial Decisions: Using

Rational Choice theory and Centrographic


techniques for Geographical Profiling












TL
[Pick the date]



Abstract

We present two mathematical models to assist in the investigation of a serial killer.
The models take the locations of previous crime sites and the order that they occurred as
variables. These data are few compared to the seemingly infinite possibilities of the
criminals actions, but at least one thing is known: the criminal will do his best to avoid
being caught. We infer then, in accordance with past research, that he will kill where he
knows the surroundings. This motivates the assumption that the killer is more likely to kill
closer to his home or places he has killed before. We further assume that the future
locations of attacks depend only on the location of the killers home and the location and
relative time of past crimes.
The first of the two models, the centrographic model, builds upon earlier work
by LeBeau. It uses an analogy to kinematics to derive the location and movement of the
centroid of a series of crime sites. We consider the change in the location of the
centroid over time to predict future crime site locations. In other words, we model the
series to have a velocity and momentum and use this information to predict the
future location of the centroid.
Our second model, the Rational Serial Criminal model, uses methods from
microeconomic theory to model the criminal as a rational agent. This model formalizes
the process of looking at a series of crime locations and assessing where the criminal prefers
to find victims or dump bodies. Using data on a series of crimes, the model outputs
contoured surfaces representing areas where the killer is more or less likely to live. From
here, we extend the model to predict where a killer is likely to strike next.
We run simulations on crime site data for different serial killers and compare our
location predictions to the actual next kill in the series. The centrographic model performs
well as measured by the error distance. Un- fortunately, the model is sensitive to a parameter
that must be determined for each individual series. The rational choice model also
performs well predicting home location as measured by the Hit Score %, a metric which
compares the efficiency of a search informed by prediction to a random search.
Introduction
A serial killer is, traditionally, a person who has murdered three or more people

over a period of more than a month, with down time (a "cooling off period") between the
murders. Some sources, such as the FBI, disregard the "three or more" criterion and
define the term as "a series of two or more murders, committed as separate events,
usually, but not always, by one offender acting alone" or, including the vital
characteristics, a minimum of two murders
When a serial killer is on the prowl , the police deploy every available resource to
stop the killings and bring the murderer to justice. Serial killers provide an interesting
challenge to geographic profilers, due to the relative paucity of data on these criminals and
their unique geographic behavior. Statistical prediction methods in widespread use for
other types of serial crimes such as burglaries [8] or auto thefts [39] are less useful for
finding serial killers, as these methods rely on a large initial data set to generate
predictions.
Experience shows that the behavior of such serial criminals follows some
predictable patterns. Specifically, serial killers tend to operate near their homes, in areas
with which they are familiar [32]. The inclination of a serial killer to work close to a
home base or anchor point [32] often leads to a spatial clustering of their crime sites.
Therefore, by looking at the geographic locations of crimes, we can say something about
where the criminal who committed them is likely to live or might strike next. This report
is conducted in order to bring readers and police knowledge to reduce criminal rate
relating to this matter and then we also compare merits as well as shortcomings of two
models in the report.
We attempt to predict a serial killers future behavior and location data on their
crime sites and initial assumptions about their behavior. In Section 3, we adapt two
complementary models to predict future crime sites and point to the location of a serial
killers anchor point [32].
We tested our models by implementing each one in a computer simulation, and
running them on partial data sets from two historical serial killers. We describe our
data collection process and provide examples in Section 5. We compared our models
predictions for the locations of later crimes to the actual locations where later crimes in
each series occurred. We describe the results of these simulations in Section 6. Finally, in
Section 7, we note some shortcomings of our approach and propose directions for further
research.
At first blush, criminals seem difficult, if not impossible, to predict. The crimi-
nal, however, must follow some sort of thought process in order to commit his acts and, as
such, crime pattern theory combines rational choice, routine activity theory, and
environmental principles to explain the distribution of crimes [32]. From these ideas,
geographical profiling was developed. When analyzing serial crimes, few data are
available other than that at the crime scene and disposal site of the body. This motivates
Lundrigan and Canters research into a serial killers disposal site location choice, which
statistically analyzed data on apprehended serial killers to determine how a killers home
location related to his body disposal sites [23]. Many factors can influence the serial killers
disposal site selection. In a study conducted in Germany, the median distance between
the disposal site and the killers home decreased as age of killer increased, a killers
higher IQ corresponds to a further distance of disposal site from the home base, though
the number of murders and distance between a killers home and disposal site was not
statistically significant [38].
Another method of determining the location of a serial killer is to consider the
daily routines of the victims. Geographic profilers can use mapping software to
determine where victims routines overlap, and these locations can be interpreted as
places where the killer might have seen the victims. These data, in turn, can be used to
hypothesize the killers daily whereabouts [37]. In an age where most people are
connected to the rest of the world through cellular telephones and internet, it is possible
now more than ever to gather objective data regarding where a person was at a certain
time.
In modeling serial crime, a primary concern of geographical profiling is the
prediction of areas at high risk of future crime, known colloquially as crime hotspots.
Past models for geographical profiling of serial offenders have shown varying results. Due
to the nature of serial crimes as discrete events, the ma- jority of methods for displaying
and analyzing data on serial criminals utilize a technique known as point mapping [10]. A
problem with simple point mapping, however, as described by Chainey et al., is that it is
difficult to clearly identify the location, relative scale, size and shape of hotspots when
crime data are presented as points. As such, it becomes necessary to find alternate
methods of display and analysis.
Literature Review:
A limited amount of information is available regarding serial murderers spatial
decisions. Existing research on this topic has been based primarily on American serial
murderers, and only a few studies have quantified the distance between where serial
murderers live and where they offend (Canter, Missen, & Hodge, 2000; Godwin &
Canter, 1997; Hickey, 1991; Lundrigan & Canter, 2001; Rossmo, 2000). Moreover, most
of these studies did not consider the factors that potentially influence this violent form of
spatial decision-making. Given that police investigators sometimes use geographic
profiling techniques to predict offender home locations (Canter, Coffey, Huntley, &
Missen, 2000; Rossmo, 2000), a better understanding of the relationship between
offender characteristics and spatial decisions could enhance the predictive accuracy of
such endeavours.
Methodology
To absolutely and thoroughly understand two models for geographical Profiling :
Centrogaphic Model and Rational Serial Criminal Model, we have read a wide range of
professional articles such as Geographic profiling analysis: principles, methods, and
applications, in Crime Mapping Case Studies of two famous writers in this field:
D. K. Rossmo and L. Velarde or An application of geographic information
systems (GIS) : the utility of victim activity spaces in the geographic profiling of
serial killers of C. Shamblin at Louisiana State University. After that, members of the
team synthesize to give viewers and readers accessibly comprehensive information:
Centrographic Model
1. Background:
A standard analytical measure in the geographical profiling toolbox is the spa- tial
mean of point data, obtained by summing the geographic coordinates over all observations
independently in each spatial dimension and dividing by the total number of observations
[32]. This effectively generates a center to the data set that can be used in conjunction
with other methods to describe the concen- tration of a distribution. A long-utilized and
relatively simple tool, the spatial mean remains in widespread use today in geographical
profiling due to its ease of use. Despite the simplicity of calculation, previous studies
have shown the spatial mean to perform comparably to alternative, more complex
algorithms and even human judgment [39]. One drawback to the spatial mean is that it
doesnt intrinsically use time to describe and forecast an evolving distribution such as is
seen with serial killings. To utilize the temporal aspect of such a data set, we must extend the
model. One extension that arises naturally is to consider the change in the spatial mean
over time [21]. Previous work has used analogy to simple physical phenomenon to come
up with parallels to velocity (change in position over time), acceleration (change in
velocity over time), and momentum (the product of velocity and mass) [32, 21]. In this
model we will use several of these concepts to come up with a method of predicting the
locations of future kills or crimes in a spree.
2. Assumpsions:
To justify a centrographic approach to geographical profiling, a few assumptions
must be made regarding the behavior of serial killers:
A serial killer is more likely to kill in an area hes familiar
with, (i. e., a place he lives near or has killed near before.)
The distribution of probable locations of next kill changes with
time, even if the distribution of known kills does not.
The order in which and time at which kill events in a
distribution occur are a factor in determining probable
locations of next kill, not just the spatial location.